4.0.0

• reranking added
• separate LLM, embedding models and reranking models
• new services (reranking and detokenize)
• models for reranking and embeddings added
• vicuna promopt added
• llama namespace removed from LlamaClientNode
• full demo with LLM + chat template + RAG + reranking + stream
• README:
  • model shards example added
  • reranking langchain and demo added
  • embedding demo added
  • minor fixes
• langchain reranking added
• langchain upgraded to 0.3
• llama.cpp b3870

3.9.2

• chat_llama_ros added to README
• model shard files download added
• llama.cpp b3827

3.9.1

• qwen2 updated to qwen2.5
• llama.cpp b3799

3.9.0

• new sampling from llama.cpp
• grammar functions removed
• n_remain removed
• threadpool added
• llama.cpp b3756

3.8.3

• fixed stop when n_remain is 0
• llama.cpp updated

3.8.2

• ChatLlamaROS stream fix
• ChatLlamaROS demo video added
• Fix passing image as data

3.8.1

• llama.cpp updated
• new cpuparams

3.8.0

• service to list LoRAs added
• service to modify LoRAs scale added
• Qwen2 added
• Phi-3 repos fixed
• llama.cpp updated

3.7.0

• Format chat messages service create
• llama_ros_common for langchain integration
• Langchain Chat integration
• new chatllama_demo create
• n_threads_batch default set to 1
• llama.cpp update

3.6.0

• Llama and Llava destructors fixed
• gpt_params renamed to llama_params
• params are now structs managed by functions instead of a class
• llama.cpp updated